Mamba
PulseAugur coverage of Mamba — every cluster mentioning Mamba across labs, papers, and developer communities, ranked by signal.
20 day(s) with sentiment data
-
New D2Mamba Model Excels at Segmenting Tiny Retinal Lesions
Researchers have developed D$^2$Mamba, a novel deep learning model designed for segmenting small and irregularly shaped lesions in retinal images, specifically for diabetic retinopathy. This model integrates non-causal …
-
New foundation model SepsisFlow enhances ICU sepsis risk prediction
Researchers have developed SepsisFlow, a new foundation model designed to provide continuous risk assessment and trajectory modeling for sepsis in intensive care units. This model integrates a Mamba-based foundation mod…
-
Researcher integrates Forth interpreter into Nemotron 3.5 Lightning model weights
A researcher has integrated a Forth interpreter directly into the weights of NVIDIA's Nemotron 3.5 Lightning model, enabling it to execute Forth code as part of its generation process. This novel approach, inspired by t…
-
New State Space Model Architecture Enhances Image Deblurring
Researchers have introduced XYScanNet, a novel architecture for single image deblurring that utilizes a State Space Model (SSM) approach. Unlike previous Mamba-based methods that convert image patches into 1D sequences,…
-
OcTreeMamba framework enhances 3D medical segmentation with Octree and Mamba
Researchers have introduced OcTreeMamba, a novel framework designed for 3D medical image segmentation. This approach integrates Octree structures with State Space Models (SSMs) to enhance the propagation and integration…
-
New framework reveals first token's 'attraction' in Mamba models
Researchers have developed a new analytical framework to understand the dynamics of deep Mamba models, focusing on the influential role of the first token. Unlike attention mechanisms, Mamba models use matrix-valued int…
-
MambaTG framework enhances temporal graph link prediction using Mamba architecture
Researchers have introduced MambaTG, a novel framework designed for future link prediction on temporal graphs. This model addresses the challenge of predicting both recurring and entirely new connections by focusing on …
-
SSFMamba framework enhances 3D medical image segmentation using symmetry and frequency modeling
Researchers have developed SSFMamba, a novel framework for 3D medical image segmentation that integrates spatial and frequency domain modeling. This approach, based on Mamba, incorporates Hermitian symmetry and a Multi-…
-
New UINO-FSS framework unifies foundation models for few-shot segmentation
Researchers have developed UINO-FSS, a novel framework for few-shot semantic segmentation that unifies knowledge from multiple foundation models into a single-encoder architecture. This approach addresses the limitation…
-
New distillation method transfers Transformer knowledge to Mamba models
Researchers have developed a new method for distilling knowledge from Transformer models into State Space Models (SSMs) like Mamba. This two-stage approach involves first distilling knowledge into a linearized Attention…
-
Apple researchers distill Transformer knowledge into Mamba architecture
Researchers from Apple Inc. have developed a new method for distilling knowledge from Transformer models into Mamba-like architectures. Their two-stage approach involves first distilling knowledge into a linearized Atte…
-
Near-zero embeddings stall gated model training; new fixes proposed
A new paper identifies a critical training instability in gated sequence models, particularly those with untied embeddings. This issue, caused by near-zero input embeddings, can lead to extreme gradient amplification, e…
-
New Mamba-based model forecasts consumer preferences with fuzzy logic
Researchers have developed a fuzzy selective state model to forecast consumer preferences, utilizing a Mamba-type architecture. This model describes customer preferences as triangular fuzzy numbers, with their evolution…
-
Dynamical Memory Transformer offers constant cost attention for long sequences
Researchers have introduced the Dynamical Memory Transformer (DMT), a novel hybrid architecture designed to improve how models handle long sequences. Unlike standard transformers that store an explicit history, leading …
-
Seismic Foundation Models Questioned as Feature Extractors
A new research paper questions the efficacy of current foundation models (FMs) in seismology, arguing they function more as glorified feature extractors than true FMs. The study, which experimented with nine Mamba-based…
-
NVIDIA releases Nemotron 3.5 Transcribe, a 40-language open speech LLM
NVIDIA has released Nemotron 3.5 Transcribe, an open-weights speech large language model capable of transcribing audio in 40 languages. This model, featuring approximately 4.3 billion active parameters and a 65,536-toke…
-
New Bi-Mamba Model Enhances Robot Navigation with Real-Time Trajectory Prediction
Researchers have developed TrajMamba, a novel bidirectional state space model designed for real-time human trajectory prediction using RGB-D cameras. This model addresses the train-deploy pipeline gap by aligning pose e…
-
New MDF-Mamba Model Enhances MODIS Spectral-Temporal Unmixing
Researchers have developed a novel Multi-task Disentangling Fusion Mamba (MDF-Mamba) model to improve spectral-temporal unmixing of Moderate-Resolution Imaging Spectroradiometer (MODIS) time series data. This model addr…
-
LayoutMamba combines Transformer and Mamba for efficient document analysis
Researchers have introduced LayoutMamba, a novel framework for document layout analysis that combines sparse-attention Transformers with Mamba's state-space mechanism. This hybrid approach aims to balance detailed local…
-
New research explores efficient fine-tuning and compression for vision foundation models
Two recent arXiv papers explore methods for making large vision foundation models more efficient. The first paper, by Niclas Hergenröther, investigates data-efficient fine-tuning techniques like Visual Prompting (VP), L…